Table 1.
Formulas of several commonly used vegetation indices.
Figure 1.
Areas in which LAI was measured.
The effectiveness of the newly developed moisture adjusted vegetation index (MAVI) is tested using LAI measured in two grassland and two forest study areas in China: (A) Tiantongshan Mountain forest, (B) Maoershan Mountain forest, (C) Hulunbeier grassland, (D) Xinlinhaote grassland. The images composed of reflectance in bands 4, 3 and 2 from Landsat-5 TM are also shown.
Table 2.
The information on the study areas and the Landsat-5 TM images used in this study.
Figure 2.
The best fitted relationships between LAI and vegetation indices.
The MAVI and three soil adjusted vegetation indices (SAVI, OSAVI, and MSAVI) are compared in the four study areas: (A) Tiantongshan, (B) Maoershan, (C) Hulunbeier, and (D) Xilinhaote. The statistics of the best fitted VI-LAI relationships are listed in Table 3. MAVI produces a higher R2, smaller normalized RMSE of retrieved LAI compared with the three soil adjusted vegetation index in both forest and grassland areas.
Figure 3.
The best fitted relationships between LAI and vegetation indices.
The MAVI and three selected vegetation indices (NDVI, MNDVI, and RSR) are compared in the four study areas: (A) Tiantongshan, (B) Maoershan, (C) Hulunbeier, and (D) Xilinhaote. MAVI produces the largest R2 and the smallest normalized RMSE of estimated LAI in Tiantongshan and Hulunbeier. The performance of MAVI based on R2 and RMSE is only slightly second to RSR in Maoershan and Xilinhaote. These results prove that MAVI has stable correlations with LAI under different cover types through incorporating the SWIR reflectance in band-ratio form.
Table 3.
The best fitted VI-LAI relationships and their statistics.
Figure 4.
Canopy reflectance of Jack Pine and Black Spruce forests against LAI for different backgrounds.
The canopy reflectance of Jack Pine and Black Spruce forests is simulated using the 4-Scale model against LAI for different backgrounds (moss, lichen, and forest soil). A synthetic background consisting of 50% of water and 50% of moss is also included for the Black Spruce forest [19]. The sensitivity of MAVI to background reflectance disturbances is investigated using these modeled results.
Figure 5.
Background reflectance effects on vegetation indices at different LAI values in Jack Pine forest.
The effects of different backgrounds (moss, lichen, and forest soil) on the selected vegetation indices (SAVI, OSAVI, MSAVI, MNDVI, RSR, MAVI, NDVI, and SR) are simulated using the 4-Scale model for the different LAI levels in Jack Pine forest. The forest background strongly affects the values of SAVI, OSAVI, MSAVI, and NDVI as the LAI values are less than 2. MAVI and SR can reduce the effects of forest backgrounds at low LAI values. RSR and MNDVI show the smallest background reflectance effects among these vegetation indices.
Figure 6.
Background reflectance effects on vegetation indices at different LAI values in Black Spruce forest.
The effects of different backgrounds (moss, lichen, forest soil, and the mixed background of water and moss) on the selected vegetation indices (SAVI, OSAVI, MSAVI, MNDVI, RSR, MAVI, NDVI, and SR) are simulated using the 4-Scale model for the different LAI levels in Black Spruce forest. The results are similar to those of Figure 5, but RSR and MNDVI do not perform much better than other vegetation indices in reducing the effect of the mixed background of water and moss.
Figure 7.
Sensitivity of different vegetation indices to forest background reflectance.
The background reflectance strongly affects the values of SAVI, OSAVI, MSAVI, and NDVI in both Jack Pine and Black Spruce forests as the LAI values are less than 2, leading to larger TVI values of them compared with other vegetation indices. RSR and MNDVI present small TVI values at low LAI values, indicating that the background reflectance effects on them are smaller than other vegetation indices. MAVI has relatively small TVI values over the entire LAI ranges in both Jack Pine and Black Spruce forests. The results demonstrate that MAVI that combines the red, NIR and SWIR reflectance can reduce the effects of background reflectance on forest canopy LAI retrieval.
Figure 8.
Variations of different vegetation indices with aspects in 5-degree slope intervals in Maoershan Mountain forest.
The polar angle represents aspect, and the radius represents the mean values of each VIs at a given aspect on different slopes. Topography strongly affects RSR and MNDVI, resulting in negative biases on sun-facing slopes and positive biases on sun-backing slopes. The values of SR, NDVI, and MAVI increase as the slope increases, which is similar to the changes of forest age with slope. It can be inferred that vegetation indices expressed in band-ratio form are able to remove a large proportion of topographical noise.
Figure 9.
Effects of slope variations on different vegetation indices.
Note: (A) the coefficient of variation (CV) of each vegetation index varies with slopes, (B) the R2 values of linear correlations between vegetation indices and the cosine of the incidence angle vary with slopes. The CV values of MNDVI vary from 5.32% to 13.02% corresponding to the slopes from 5° to 25°, which shows the largest topographical noise among all the selected vegetation indices. RSR has the second largest CV values ranging from 7.09% to 9.81%. SR presents a medium CV values in the range from 3.14% to 5.37%. The CV values are quite small for NDVI and MAVI ranging from 0.64% to 1.92% and from 0.80% to 2.56%, respectively, implying that NDVI and MAVI can remove much of topographic noise through expressing in band-ratio form. The conclusions based on R2 are also similar.